DOE/OEDI COA bulk density and API qualification

This qualification advances refinery issue #3305 from one selected crude to the complete usable matrix in the public DOE/OEDI summary workbook. It tests whether OilAssayCharacterisation reconstructs bulk specific gravity and API gravity from complete liquid-volume-basis category tables without inventing terminal boiling properties or mutating the thermodynamic system.

Public source, license, and method

The source is the U.S. Department of Energy/National Renewable Energy Laboratory Crude Oil Analysis (COA) Database, distributed through the Open Energy Data Initiative (OEDI):

The archive contains Summary of Analyses.xls and COAMDATA_DESC.pdf. The database compiles 9,076 crude-oil analyses made by the U.S. Bureau of Mines and later maintained by DOE. The method description records standardized atmospheric and 40 mm Hg distillation, fraction-volume and gravity measurements, and refinery category aggregation.

The regression selects all workbook rows that satisfy these predeclared completeness rules:

  1. gasoline+naphtha, kerosene, gas-oil and residuum yields sum to 100.0 vol%;
  2. every positive-yield category has a reported specific gravity;
  3. nonviscous, medium and viscous lubricating-distillate yields are zero;
  4. the row reports whole-crude specific gravity and API gravity.

Exactly five rows qualify. No row is removed because its reconstructed error is unfavorable.

Frozen qualification matrix

Sample Location/field Gasoline+naphtha vol% / SG Kerosene vol% / SG Gas oil vol% / SG Residuum vol% / SG Published crude SG / API Reconstructed SG / API Absolute SG / API error
920 Turner Valley, Alberta 70.5 / 0.754 13.2 / 0.813 3.7 / 0.831 12.6 / 0.889 0.779 / 50.1 0.781647 / 49.528 0.002647 / 0.572
50146 Ranch W, Texas 17.8 / 0.794 0.0 / — 78.7 / 0.856 3.5 / 0.907 0.847 / 35.6 0.846749 / 35.610 0.000251 / 0.010
56337 Manderson, Wyoming 88.7 / 0.767 8.7 / 0.794 0.0 / — 2.6 / 0.815 0.771 / 52.0 0.770597 / 52.124 0.000403 / 0.124
60205 South McCallum, Colorado 75.0 / 0.741 19.8 / 0.804 4.2 / 0.842 1.0 / 0.873 0.765 / 53.5 0.759036 / 54.921 0.005964 / 1.421
68120 Vermilion Block 14, Louisiana 49.7 / 0.749 38.5 / 0.805 0.0 / — 11.8 / 0.832 0.782 / 49.4 0.780354 / 49.828 0.001646 / 0.428

Published category and crude values are reproduced at their source precision. A zero-yield category is omitted from the Java assay rather than assigned a fictitious density.

Calculation and acceptance

For resolved mass fractions w_i and cut specific gravities SG_i, NeqSim applies ideal additive liquid volumes:

\[SG_{bulk}=\left(\sum_i\frac{w_i}{SG_i}\right)^{-1}\]

For a normalized liquid-volume-basis table this reduces to:

\[SG_{bulk}=\sum_i v_iSG_i\]

The corresponding API gravity is:

\[API=\frac{141.5}{SG_{bulk}}-131.5\]

The regression requires:

Across all five qualifying rows, the observed SG absolute-error maximum/mean/RMSE is 0.005964 / 0.002182 / 0.003017. The API absolute-error maximum/mean/RMSE is 1.420670 / 0.510845 / 0.713312 degrees API. These errors are frozen as evidence and are not tuned away.

Maturity, validity, and stop boundary

Within the frozen COA matrix, the ideal-additive-volume behavior is qualified for assay screening over published whole-crude SG 0.765–0.847. This is not a blend-density standard, custody-transfer calculation or design certification.

The calculation uses the density reference condition represented by the source table. It does not model temperature or pressure correction, excess volume, blend contraction, sulfur/heteroatom effects, or uncertainty correlations between reported fields. The largest error is retained and limits the stated accuracy.

The matrix intentionally does not create pseudo-components because the summary categories do not provide finite representative boiling points and molar masses for all terminal categories. This increment adds no production formula, coefficient tuning, terminal-cut extrapolation, TBP/ASTM conversion, column change, JSON/MCP schema, notebook, vacuum model, blending optimizer or conversion-unit model.

Python uses the same authoritative Java methods through the normal NeqSim JVM gateway; no separate Python property equation is maintained.

Physical density at 60 degF is available separately through getBulkDensityKgPerCubicMetreAt60F(), using 999.016 kg/m3 for water. API-gravity inputs remain dimensionless SG60/60 values and are not pre-multiplied by water density; this preserves exact API-to-SG round-tripping while keeping physical-density units explicit.